As screenwriters and studios negotiate AI’s role in the entertainment industry, it’s important to be mindful of some core copyright protection principles.
In a decision that surprised exactly no one, D.C. District Court Judge Beryl A. Howell ruled last Friday that the Register of Copyrights did not act “arbitrarily or capriciously” in denying a copyright registration to Dr. Stephen Thaler for artwork generated entirely by artificial intelligence. “Human authorship,” Judge Howell wrote, “is a bedrock requirement of copyright,” which has never stretched so far as to protect works generated “absent any guiding human hand.”
The Copyright Office’s position on this subject has always been crystal clear: works created by “non-humans”—whether they be AI, animals, or divine beings—aren’t eligible for copyright protection. But while human involvement is essential to copyright protection, the mere presence of a human author doesn’t provide some secret backdoor way of making AI-generated content copyrightable. Repeat after me: content created by human beings can be protected by copyright. Content generated by AI can’t.
This distinction may have have been lost on some observers in the days following the Thaler decision—especially against the backdrop of the ongoing writers’ strike, which is about to enter its fifth month. An article in the Hollywood Reporter earlier this week suggested that there’s finally been some movement between the parties with respect to generative AI, as studios recognize that copyright protection in AI-generated scripts is only possible for those works if they’re revised by human writers. The article quoted an unnamed source close to the AMPTP as saying,“If a human touches material created by generative AI, then the typical copyright protections will kick in.”
Not exactly. This may be true, but only if we recognize that the “typical copyright protections” don’t include AI-generated content which makes up an appreciable part of a work. If content contributed by AI is more than de minimis and would qualify for copyright protection had it been created by a human author, that content needs to be disclosed to the Copyright Office and will be excluded from protection.
Welcome to the world of “unclaimable material,” a strange land where material that could be protected by copyright isn’t protected by copyright.
What Is Unclaimable Material?
For purposes of a copyright registration, “unclaimable material” has historically included four types of material:
- Previously published material.
- Previously registered material.
- Material that is in the public domain.
- Copyrightable material that is owned by a third party.
In March of this year, the Copyright Office issued a new policy statement that essentially added a fifth category of unclaimable material to the list: AI-generated content. If AI technology determines the expressive elements of its output, the generated material is not considered to be the product of human authorship, and therefore not protected by copyright.
What’s important to remember is that if a work contains an “appreciable amount” of unclaimable material, that material needs to be disclosed and disclaimed in a copyright registration application. The disclaimed material is then excluded from the claim of copyright.
While Dr. Thaler attempted to register a copyright claim in material created entirely by AI, it’s more often the case that a work containing AI-generated material will also contain sufficient human authorship to support a copyright claim. For example, a human author may select or arrange AI-generated material in a sufficiently creative way. Or an author may modify material originally generated by AI to such a degree that those modifications meet the standard for copyright protection. In these cases, copyright will protect the human-authored aspects of the work. Importantly, however, there will still be no copyright protection in the AI-generated material itself.
Think of human modifications to AI as a quasi-derivative work—the copyright in a derivative work only extends to the material contributed by the author of that work, as opposed to the underlying material. “The copyright in such [derivative] work is independent of, and does not affect or enlarge the scope, duration, ownership, or subsistence of, any copyright protection in the preexisting material.”
An example of this rule in action is demonstrated by the recent Copyright Office registration decision involving Kristina Kashtanova’s Zarya of the Dawn. The Office concluded that the original text and compilation of the elements in Kashtanova’s comic book were sufficient to support a claim of copyright. However, it canceled their original registration for failing to disclose that the comic’s illustrations were generated by AI. A newly-issued registration clarifies that copyright protection in Zarya does not extend to any of the individual images themselves because they lack the requisite level of human authorship.
“Appreciable” vs. “De Minimis” Amounts of AI-Generated Content
While AI-generated material will never itself be protected by copyright, it only needs to be disclosed and excluded from a copyright application if a work contains an appreciable amount of AI-generated content. So-called “de minimis“ uses of AI don’t need to be disclosed. Determining where and how to draw the line between appreciable and de minimis requires conducting a counterfactual exercise: Would the AI-generated material be copyrightable if it had been created by a human author? If the answer is yes, the Copyright Office considers it appreciable and disclosure is required.
Many times, the answer will be no. If a work contains material that is uncopyrightable, such as facts, short phrases or mere ideas, there is no need to exclude that material from a claim of copyright—regardless of whether it was generated by a human or by AI. So, for example, if an author uses AI for brainstorming and ideation (offering potential character names, ideas for plot points, etc.), but no appreciable amount of AI-generated material is contained in the final script submitted for registration, there’s no need to disclose or exclude the AI. The same is true if AI is used for technical tasks—blurring out license plate numbers from reality show footage or performing color corrections. These tasks, if performed by humans, wouldn’t be considered copyrightable contributions, so they don’t need to be excluded if performed by AI.
On the other hand, if a director uses AI to produce some of the special effects for a movie, the AI-generated content would need to be disclaimed, because they would be considered a copyrightable contribution to the motion picture if the material had been created by a human author. The same is true if a writer were to use ChatGPT to generate a rough draft of a script, which he then lightly polishes and punches up. The resulting screenplay is eligible for copyright protection, but so long as the finished product still retains underlying AI-generated material that would also be considered copyrightable, that material needs to be disclosed and is excluded from the claim. The mere presence of a “human touch” doesn’t confer copyright protection on the AI-generated portions.
What Happens if You Don’t Disclose Unclaimable Material?
The Copyright Office’s disclosure requirements aren’t onerous, and disclaimers of unclaimed material can be quite general. Don’t worry about providing a detailed spreadsheet of every use of AI technologies in the creation of your work. It’s typically enough to include a simple statement in the excluded material section of the copyright application to the effect that “some text was generated by AI.” If the Copyright Office has more specific questions, they’ll ask.
Occasionally, the Register of Copyrights is asked to provide advice to courts on whether the Register would have refused registration had the Copyright Office known of alleged inaccuracies in a copyright application. For example, just this week, the Copyright Office concluded that a tattoo design affixed to LeBron James’ chest was entitled to registration even though it’s based on a modified version of the Venetian Resort logo. The Office noted that, had a registration specialist known that the Lion tattoo design contained an appreciable amount of material owned by a third party, the specialist would have asked the artist to disclose and exclude that material from his claim.
You may be asking yourself, how would anyone know whether or not you used AI to generate dialogue for your screenplay? Unlike pre-existing works in the public domain, it’s nearly impossible to accurately prove that a work has been created by AI—at least for now. In the future, there may well be tools that can determine whether a particular work is comprised of an appreciable amount of AI-generated material. By not disclosing, you run the risk of having your entire copyright registration invalidated.
Contrary to popular belief, I don’t think it’s realistic to think that studios will replace screenwriters with movie and television scripts churned out by ChatGPT. Strike or no strike, no one wants to spend millions of dollars to produce and market a work that they can’t fully protect. And merely having a human writer give minor touch ups to an AI-generated screenplay won’t provide the desired copyright protection either.
The greater the amount of AI material that finds its way into a final product, the less protection that output will receive. Therefore, it seems much more likely that studios are going to want to minimize the amount of off-the-shelf AI incorporated into their creative output, at least insofar as that material would be entitled to copyright protection had it not been created by AI. That will require writers who identify as human.
As always, I’d love to know what you think. Let me know in the comments below or @copyrightlately on whatever social media app you aren’t boycotting this week.